7. Building FENT Nanopore Transistor Chip Prototype
preemptdisease.substack.com
Building any new kind of nano-scale CMOS transistor is an uphill task, generally speaking. Fabricating ‘in silicon’ for the first time is capital intensive, a fact semiconductor industry insiders know quite well. That’s the reason the semiconductor industry invests heavily on TCAD and EDA computational tools to numerically simulate new transistors and chip architectures, to fully verify and mitigate risk, well in advance of beginning prototype fabrication.
In the case of FENT (Field Effect Nanopore Transistor), there was added complexity. It is an entirely new transistor design, where the silicon channel is vertical and the source/drain are on top and bottom (rather than in a horizontal plane). In addition, we have a nanopore incorporated through the transistor channel from top-side to bottom-side.
While our NIH grant got us started, we quickly realized building the device itself was quite costly. We were able to develop and demonstrate a new method of fabricating sub-5 nm nanopores in silicon (data not published). And it also enabled us to perform numerical simulations to understand and verify FENT transistor device function. Raising investor capital was a chicken-and-egg problem, so we paused on the FENT project for a while. And in the interim, focused on our second proteomic platform (will inform on this soon).
We got a break on our FENT project in 2019. I had a chance meeting with DARPA Program Manager Dr. Eric Van Gieson at a conference in San Francisco. I think it’s fair to say that my excited pitch for a fully-CMOS nanopore technology for point-of-need sequencing got his attention. Following this, DARPA Biological Technologies Office (BTO) funded INanoBio under its ECHO program to build the FENT prototype, as part of a larger team led by Icahn School of Medicine at Mount Sinai, New York.
Deeptech projects like FENT are inherently high-risk, high-reward and need a significant amount of capital to prove viability. The contrast is night and day when compared to new NGS technologies that can readily generate proof-of-concept sequencing data with just seed capital. Complementing DARPA funding of FENT, our angel investor group provided over 3X in matching funds to support successful completion of FENT nanopore transistor prototype fabrication.
We successfully made the first FENT prototypes within 3 years of start of the fabrication effort (launched in 2019). Even with a nanopore incorporated into the silicon channel - FENT works as an excellent transistor. Below, Mukil holding the ~1 cm x 1 cm FENT prototype chip.
Windows or wells are fabricated on the backside of the chip (handle silicon) to connect to the nanopore incorporated in the FENT transistor on the top silicon film of SOI wafer. We successfully made FENT with 50 nm nanopores and demonstrated FENT cut-off frequency at >100 MHz (capable of over 100 million samplings per second). Next steps are to reduce the diameter of nanopore to 5nm and demonstrate proof of concept ultra-fast DNA sequencing. We have started testing on FENT electron-current sensing of DNA, nanoparticles passing through the pore, and safe to say, are planning a peer reviewed publication sometime next year.
Being fully solid-state CMOS based technology, I believe the FENT nanopore platform will be one generation ahead of all protein nanopores and ion-current sensing based nanopore sequencers. The ability to readout bases at speeds of up to million bases per second (per pore) and scale FENT chip manufacturing at established semiconductor manufacturing foundries is at the core of this assertion.
Let’s discuss with a few example numbers - that’s when it becomes fun and illustrates the absolutely exciting possibility space with FENT technology.
The size of the FENT transistor itself is < 2 micron. Current FENT Gen1 prototype chips have a test array with FENT devices spaced around 15 microns apart. However, as we discussed previously in post #6, section 2, the complexity of on-chip and off-chip signal sensing electronics (ASIC) define the array size and density of nanopore sequencing devices, generally speaking. This in-turn depends on the exact DNA sequencing approach and chemistry. With FENT we will be exploring multiple sequencing approaches.
I estimate that first generation commercial FENT sequencing chips will have the capability to read around 100 million bases per second, per sq cm chip. So, we should be able to read the 3 billion base pair human genome with 15x coverage in under 15 minutes. This is a high-level order of magnitude estimate for initial FENT sequencing products. At this throughput, FENT should be competitive against all NGS sequencing platforms and protein nanopore sequencers, projected to 2027.
If you noticed, I have started referencing the sequencing speed not only with respect to time (per second), but also against the area of the chip. This is because I believe small form-factor sequencing devices and one-time-use disposable sequencing chips are going to be the defining criteria and drivers for wide adoption and democratization of sequencing-based clinical diagnostics, at the clinic and at the point-of-care.
Most clinical diagnostics are based on targeted sequencing and do not need whole genome sequencing. These less complex applications might require FENT chips that are around 10 sq mm, with smaller FENT array. 10 sq mm FENT chip ~ 3.3 mm x 3.3 mm chip area (for disambiguation). On the other hand, complex applications like cancer tumor profiling and single cell sequencing would require multiple sq cm FENT chips. So on the average we can assume each disposable FENT chip to have an area of around1 sq cm.
Historically, total DNA sequencing volumes have grown by ~ 50% CAGR (compound annual growth rate) due to continued lowering of sequencing costs. Extending this growth rate, Ark Invest Research estimates the total sequencing volume in 2024 to be 105 million whole human genome equivalents. INanoBio expects this growth rate to continue (and even increase) over the next decade, driven in part by rapid drop in sequencing costs and ever-increasing clinical applications of sequencing, to reach around 1.8 billion annual human genome equivalents in 2031.
How many 300mm FENT wafers would we need to address a significant fraction of this market? To estimate this we performed a scenario analysis (BOP - base, optimistic and pessimistic-case), see below. We assumed that our FENT sequencer may capture 2.5 - 3.75% of sequencing market share in 2027, which would increase to 7.5 - 25% by 2031. Based on the above projected total annual sequencing volumes, we arrive at around ~ 300,000 300mm FENT wafers / year needed to supply the projected 16.5% base-case FENT sequencing market share in 2031. This rate of manufacturing will produce around 150 million 1 sq cm FENT chips per year, each capable of sequencing one whole genome equivalent.
So, a single large 300mm semiconductor foundry can scale and address a majority of projected FENT manufacturing demand over the next decade. We can readily yield around 150 million FENT sequencing chips annually at the beginning of next decade, each capable of sequencing a whole genome in few minutes.
This illustrates the inherent advantages of scaled semiconductor manufacturing, that dramatically drives down the cost per FENT chip (will discuss in a future post).
7. Building FENT Nanopore Transistor Chip Prototype
7. Building FENT Nanopore Transistor Chip Prototype
7. Building FENT Nanopore Transistor Chip Prototype
Building any new kind of nano-scale CMOS transistor is an uphill task, generally speaking. Fabricating ‘in silicon’ for the first time is capital intensive, a fact semiconductor industry insiders know quite well. That’s the reason the semiconductor industry invests heavily on TCAD and EDA computational tools to numerically simulate new transistors and chip architectures, to fully verify and mitigate risk, well in advance of beginning prototype fabrication.
While our NIH grant got us started, we quickly realized building the device itself was quite costly. We were able to develop and demonstrate a new method of fabricating sub-5 nm nanopores in silicon (data not published). And it also enabled us to perform numerical simulations to understand and verify FENT transistor device function. Raising investor capital was a chicken-and-egg problem, so we paused on the FENT project for a while. And in the interim, focused on our second proteomic platform (will inform on this soon).
Deeptech projects like FENT are inherently high-risk, high-reward and need a significant amount of capital to prove viability. The contrast is night and day when compared to new NGS technologies that can readily generate proof-of-concept sequencing data with just seed capital. Complementing DARPA funding of FENT, our angel investor group provided over 3X in matching funds to support successful completion of FENT nanopore transistor prototype fabrication.
Windows or wells are fabricated on the backside of the chip (handle silicon) to connect to the nanopore incorporated in the FENT transistor on the top silicon film of SOI wafer. We successfully made FENT with 50 nm nanopores and demonstrated FENT cut-off frequency at >100 MHz (capable of over 100 million samplings per second). Next steps are to reduce the diameter of nanopore to 5nm and demonstrate proof of concept ultra-fast DNA sequencing. We have started testing on FENT electron-current sensing of DNA, nanoparticles passing through the pore, and safe to say, are planning a peer reviewed publication sometime next year.
Being fully solid-state CMOS based technology, I believe the FENT nanopore platform will be one generation ahead of all protein nanopores and ion-current sensing based nanopore sequencers. The ability to readout bases at speeds of up to million bases per second (per pore) and scale FENT chip manufacturing at established semiconductor manufacturing foundries is at the core of this assertion.
The size of the FENT transistor itself is < 2 micron. Current FENT Gen1 prototype chips have a test array with FENT devices spaced around 15 microns apart. However, as we discussed previously in post #6, section 2, the complexity of on-chip and off-chip signal sensing electronics (ASIC) define the array size and density of nanopore sequencing devices, generally speaking. This in-turn depends on the exact DNA sequencing approach and chemistry. With FENT we will be exploring multiple sequencing approaches.
I estimate that first generation commercial FENT sequencing chips will have the capability to read around 100 million bases per second, per sq cm chip. So, we should be able to read the 3 billion base pair human genome with 15x coverage in under 15 minutes. This is a high-level order of magnitude estimate for initial FENT sequencing products. At this throughput, FENT should be competitive against all NGS sequencing platforms and protein nanopore sequencers, projected to 2027.
Most clinical diagnostics are based on targeted sequencing and do not need whole genome sequencing. These less complex applications might require FENT chips that are around 10 sq mm, with smaller FENT array. 10 sq mm FENT chip ~ 3.3 mm x 3.3 mm chip area (for disambiguation). On the other hand, complex applications like cancer tumor profiling and single cell sequencing would require multiple sq cm FENT chips. So on the average we can assume each disposable FENT chip to have an area of around1 sq cm.
Historically, total DNA sequencing volumes have grown by ~ 50% CAGR (compound annual growth rate) due to continued lowering of sequencing costs. Extending this growth rate, Ark Invest Research estimates the total sequencing volume in 2024 to be 105 million whole human genome equivalents. INanoBio expects this growth rate to continue (and even increase) over the next decade, driven in part by rapid drop in sequencing costs and ever-increasing clinical applications of sequencing, to reach around 1.8 billion annual human genome equivalents in 2031.
How many 300mm FENT wafers would we need to address a significant fraction of this market? To estimate this we performed a scenario analysis (BOP - base, optimistic and pessimistic-case), see below. We assumed that our FENT sequencer may capture 2.5 - 3.75% of sequencing market share in 2027, which would increase to 7.5 - 25% by 2031. Based on the above projected total annual sequencing volumes, we arrive at around ~ 300,000 300mm FENT wafers / year needed to supply the projected 16.5% base-case FENT sequencing market share in 2031. This rate of manufacturing will produce around 150 million 1 sq cm FENT chips per year, each capable of sequencing one whole genome equivalent.
This illustrates the inherent advantages of scaled semiconductor manufacturing, that dramatically drives down the cost per FENT chip (will discuss in a future post).